Visual Salience in the Change Detection Paradigm: The...

Visual Salience in the Change Detection Paradigm:The Special Role of Object Onset

Geoff G. Cole, Robert W. Kentridge, and Charles A. HeywoodUniversity of Durham

The relative efficacy with which appearance of a new object orients visual attention was investigated. Atissue is whether the visual system treats onset as being of particular importance or only 1 of a numberof stimulus events equally likely to summon attention. Using the 1-shot change detection paradigm, theauthors compared detectability of new objects with changes occurring at already present objects—luminance change, color change, and object offset. Results showed that appearance of a new object wasless susceptible to change blindness than changes that old objects could undergo. The authors alsoinvestigated whether it is onset per se that leads to enhanced detectability or onset of an objectrepresentation. Results showed that the onset advantage was eliminated for onsets that did not correspondwith the appearance of a new object. These findings suggest that the visual system is particularly sensitiveto the onset of a new object.

Over 100 years ago, Titchener (1901) stated that

given the course of development as we know it, the organism musthave attended to movement, etc., in its surroundings or have paid thepenalty of inattention with its life. The moving, the new and thesudden are all possible—even probable—sources of danger. (p. 209)

Clearly, the point Titchener was making concerns the evolutionarysignificance of detecting and attending to novelty. However, de-bate surrounds the types of stimulus event that can elicit aware-ness. At issue is whether the visual system is particularly sensitiveto the onset of a new object or whether onset is no more likely toattract attention than other stimulus events that can occur in thevisual field. More specifically, does abrupt onset have a specialrole in visual attention?

One particular line of evidence supporting the primacy of objectonset has come from the singleton task (Jonides & Yantis, 1988;Yantis, 1998; Yantis & Gibson, 1994; Yantis & Hillstrom, 1994;Yantis & Johnson, 1990; Yantis & Jonides, 1984). In this para-digm, participants make a speeded response to a target (e.g., aletter) presented among distractors, as in any standard search task.The crucial manipulation is the inclusion of a so-called featuresingleton: One of the items happens to be unique with respect to aparticular stimulus attribute. For instance, one of the items mightbe of a different size, of a different color, or associated withdifferent type of motion. Reaction time (RT) for the detection oftargets that coincide with the feature singleton is compared with

RT when the target coincides with one of the other (nonsingleton)items. An important aspect of the paradigm is that the target is nomore likely to be the singleton than is any other item. In otherwords, the singleton is task irrelevant. This ensures that it does notpay for participants to direct endogenous attention to the locationof the singleton. Any singleton that then accrues an RT benefitmay be said to have a tendency to attract attention.

Evidence that singletons attract attention in the absence of arelevant attentional set, however, is somewhat mixed. Jonides andYantis (1988) reported that both color and luminance singletonsfailed to attract attention. Similarly, Hillstrom and Yantis (1994)found that a target associated with the only motion in a display alsofailed to attract attention. However, more recently, Turatto andGalfano (2001) and Johnson, Hutchison, and Neill (2001; see alsoTheeuwes, 1994) have both reported data showing that colorsingletons can in fact orient attention. Gibson and Jiang (1998)suggested that feature singletons may have the capacity to elicitattention, but observers could habituate to the repetition of a salientevent (i.e., the singleton). They suggested that perhaps attention isdirected to the singleton for its initial (unexpected) presentationbut fails to be so directed after many presentations. The effectwould then be lost when RTs are averaged over many trials.Gibson and Jiang therefore presented surprise color singletonshalfway through experimental blocks. Although they did not ob-serve an attentional effect by a surprise singleton, Horstmann(2002) has recently reported such an effect. Furthermore, Kent-ridge, Cole, and Heywood (2003) showed that color singletonsdefined by local chromatic contrast attracted attention, but colorsingletons defined by spectral composition did not. This demon-strated the sensitivity of any color singleton effect to subtle vari-ations of the stimuli.

There is, however, one class of singleton which has consistentlyshown a tendency to attract attention: a singleton defined by itsrecent appearance, or an onset singleton. In an onset-singletonprocedure, a “placeholder” display is presented prior to the searchdisplay; letters are then created by the offsetting of camouflagingelements constituting the placeholders (see Figure 1). Thus, the

Geoff G. Cole, Robert W. Kentridge, and Charles A. Heywood, Depart-ment of Psychology, University of Durham, Durham, United Kingdom.

This work was supported by Medical Research Council GrantG0000679. We thank Paul Skarratt for organizing the use of participants,Amy Wilson for assistance with manuscript figures, and Patrick Cavanaghfor useful correspondence. We are also grateful to the many participantswho took part in the experiments.

Correspondence concerning this article should be addressed to Geoff G.Cole, Department of Psychology, University of Durham, South Road,Durham DH1 3LE, United Kingdom. E-mail: [email protected]

Journal of Experimental Psychology: Copyright 2004 by the American Psychological AssociationHuman Perception and Performance2004, Vol. 30, No. 3, 464–477

0096-1523/04/$12.00 DOI: 10.1037/0096-1523.30.3.464

464

letters are created by the transformation of already existing, or old,objects. When the old objects transform into letters, a new objectappears at a previously unoccupied location. This constitutes theonset singleton. The fact that onset singletons, unlike other typesof singleton, have a strong tendency to attract attention has led tothe idea that the appearance of a new object plays a special role inselection priority. However, Folk, Remington, and Johnston(1992) argued that what appears to be selection priority by newobjects may in fact be modulated by the current goals of theobserver. They showed that, for example, a color cue will attractattention if the target is defined by color but may not do so if it isdefined by luminance. They argued that the new object orientedattention as a feature singleton because participants adopted anattentional set for luminance onset singletons. This led Folk et al.to claim that object onset does not enjoy selection priority overother stimulus properties. Clearly, this work, together with thefindings that other singletons are also effective in prioritizingselection, directly challenges the idea that new objects have aspecial status in visual attention.

Recently, Cole, Kentridge, Gellatly, and Heywood (2003) tooka different approach to the issue of whether new objects have aspecial status in vision. Rather than using the usual feature single-ton task, they assessed the primacy of object onset using therelative new phenomenon of change blindness. Change blindnessrefers to the observation that participants often fail to detect largechanges that occur in the visual field if the retinal transient thataccompanies such changes is disrupted by other simultaneoustransients (Simons & Levin, 1997). For example, change blindnesscan be induced if a uniformly blank image occurs between theoriginal scene and a subsequent scene containing the change(Simons, 1996). The effect can also occur in the absence of a blankimage if a “mudsplash” appears across the scene at the point atwhich the change occurs (O’Regan, Rensink, & Clark, 1999). Ineffect, these competing transients mimic the effect created by eyemovement whereby information is lost during successive saccades(Jonides, Irwin, & Yantis, 1983; Rayner & Pollatsek, 1983). If theappearance of a new object does indeed play a special role invision, then one might expect onset to be more readily detected in

a change detection procedure compared with other changes. Putanother way, onset may be less susceptible to change blindness. Incontrast, if the visual system assigns no greater priority to onset,then the appearance of a new object should induce change blind-ness equal to that of other stimulus events.

In experiments reported previously, Cole et al. (2003) assessedthe onset primacy hypothesis using the change detection paradigmby comparing onsets with offsets. On each trial (except in Exper-iment 1), participants viewed a single pair of images of simpleabstract shapes (see Figure 2) in a variation of the standard flickerprocedure. That is, the one-shot method was used, whereby par-ticipants are presented with a single pair of images only. Thiscontrasts with the more common procedure whereby participantsare presented with two images that continually alternate until thechange is found. In Cole et al.’s (2003) object-onset condition, thesecond image of the pair contained one extra object compared withthe first image, whereas in the object-offset condition, the secondimage contained one object fewer. Participants were not informed

Figure 1. Typical trial sequence in the onset-singleton task. The search display is preceded by the placeholderdisplay. This example shows one new, or onset, object (the letter P) among six old objects.

Figure 2. An example of the type of stimuli used by Cole et al. (2003).

465VISUAL SALIENCE IN CHANGE DETECTION

as to the nature of the change; they were only asked to decidewhether the change occurred on the left-hand or the right-hand sideof the display. Results showed that onsets gave rise to greaterlocalization performance than did offsets. In other words, changeblindness was attenuated, relatively, in the onset condition com-pared with the offset condition. In addition to onsets being noticedmore frequently, RT to localize onsets was also reduced comparedwith RT to localize offsets. It was the finding that onsets are lesssusceptible than offsets to change blindness that motivated us tocarry out the present series of experiments.

The principal aim of the present research was to assess howeffective the appearance of a new object is in attenuating changeblindness compared with other events. More specifically, we com-pared detectability of new objects against a variety of changes thatoccurred at already present, or old, objects. As Enns, Austen, DiLollo, Rauschenberger, and Yantis (2001) have stated, “the rela-tive contributions of new objects versus changes to features inexisting objects remains an open question” (p. 1289). Using asingle pair of images in each trial (i.e., a one-shot procedure), weassessed the degree to which object onset was resistant to changeblindness relative to other changes. In a series of seven experi-ments, we compared new onset with changes in luminance,changes in color, and offset of old objects. Furthermore, weexamined whether it is onset per se that attenuates change blind-ness or whether it is the onset of a perceptual object.

Experiment 1

To hold that object onsets have a special status in visual atten-tion is to contend that new object detection occurs at the level of“object descriptions” or is “object based” (Duncan, 1984). In otherwords, onset primacy should not rely on (i.e., should be indepen-dent of) the detection of low-level object properties (e.g., lumi-nance). It follows therefore that if an old object undergoes aluminance change equivalent to that which accompanies a newobject, the new onset should still confer selection priority. Thiswould be because the luminance change occurring at the old objectdoes not correspond with the onset of an object. This issue isrelated to the question of what stimulus properties of a new objectalert the visual system to its onset or, put another way, whichmechanisms mediate the detection of a new object. The pathwaysknown to respond selectively to luminance change are a possiblemechanism. For instance, the magnocellular channel that origi-nates in the retinal ganglion cells and projects to the parietal areais known to be involved in the processing of luminance. Low-levelluminance processing would constitute an efficient method ofdetecting new onsets because new objects are invariably of adifferent luminance than their background. Indeed, an abundanceof evidence has shown that luminance increments effectively sum-mon attention (Jonides, 1981; Miller, 1989; Nakayama & Mack-eben 1989; Posner, 1980; Steinman, Steinman, & Lehmkuhle,1995; von Grunau & Faubert, 1994). A standard procedure is tovalidly or invalidly cue the location of a target by briefly changingthe luminance of an already present object. Typically, RT isreduced to valid luminance cues. Using the onset singleton task,Gellatly, Cole, and Blurton (1999) and Martin-Emerson andKramer (1997) have also reported that the greater the amount ofcompeting luminance change resulting from placeholders trans-

forming into letters, the less likely it is that attention will beoriented to the new item.

Yantis and Jonides (1984) originally suggested luminance de-tection as the mechanism responsible for the detection of newobjects. Later studies, however (Yantis & Gibson, 1994; Yantis &Hillstrom, 1994), showed that luminance change does not neces-sarily attract attention. For example, Yantis and Hillstrom foundthat letters undergoing a brief luminance increment at the onset ofthe search display do not accrue an RT benefit. Gellatly and Cole(2000) have also demonstrated that an object of equal luminancewith its background does have a tendency to attract attention.

Of particular importance to this issue is the work of Enns et al.(2001). Using variants of the onset singleton letter procedure, Ennset al. systematically compared onsets with a variety of luminancechanges that could occur at old objects. For instance, onset wascompared with different magnitudes of luminance change andluminance contrast. In their Experiment 4, Enns et al. systemati-cally varied the luminance contrasts of all items (i.e., new and oldobjects). This manipulation allowed selection priority to be quan-tified for different contrasts for new and old objects. The resultsshowed that contrast of old items needed to be much greater thanthat of new items in order for attentional capture to be equivalent.Enns et al. demonstrated the selection priority of new objects evenwhen old items underwent large luminance change. This providespowerful evidence that luminance change competes only weaklywith onset.

The purpose of Experiment 1 was to assess whether the onsetadvantage observed by Cole et al. (2003) in the change detectionprocedure relies on the detection of luminance change that accom-panies object onset. In order to examine this, we compared changedetection performance for two types of change. In one condition,we presented a number of trials in which the second image in-cluded an additional object (i.e., the new object condition). In asecond condition, a number of trials were presented in which aluminance change occurred at one of the old objects; the degree ofluminance change was equivalent to that accompanying the newobject. If the onset advantage is due to the processing of luminancechange, then an old object that undergoes an equivalent changeshould be as effectively detected as a new object.1

Method

Participants. Ten undergraduate psychology students took part in par-tial fulfillment of a course requirement.

Stimuli and apparatus. Between 16 and 24 objects of varying shape,color, and luminance were presented inside a virtual rectangle measuring8.5° in height and 10.6° in width. Each shape measured between 2.8° and0.9° in both height and width and was presented against a light purplebackground measuring 68 cd/m2 in luminance. The color coordinates ofthis background, measured in Lu’v’ color space (using a ColorCAL chro-mameter; Cambridge Research Systems Ltd., Rochester, Kent, UnitedKingdom), were 0.208 (u’) and 0.469 (v’). The objects presented in eachtrial were chosen pseudorandomly, with the restriction that they would

1 As in all of the experiments reported here, the participants’ task was toindicate whether they detected a change on the left-hand side or on theright-hand side of the display (or detected neither for Experiments 5 and 6).Technically, therefore, participants made a localization judgment ratherthan detection judgment. Hence, we refer to participants’ performance aslocalization.

466 COLE, KENTRIDGE, AND HEYWOOD

form two groups on either side of fixation, with an approximately equalnumber appearing in each group. The degree of luminance change resultingfrom an old object changing luminance was equal to that resulting from anew object appearing. The average luminance change was 30.3 cd/m2

across all trials. The old object undergoing change (i.e., changing inluminance) was chosen pseudorandomly from the display. However, noitem changed that occluded or was occluded by another object such thatapproximately one fifth of the object overlapped. This stipulation was alsothe case for the new onset; we had initial concerns of object masking(addressed systematically in Experiment 4b). The experiment was carriedout in a single, dimly lit room and was driven by a Pentium PC running at60 Hz and linked to a standard color monitor.

Design and procedure. A within-participant, single-variable (onset vs.luminance change), two-alternative forced-choice procedure was used. Thetrial sequence is shown in Figure 3. Each trial began with the presentationof a fixation point for 1,000 ms before the appearance of the first image for1,200 ms. A blank frame of 100 ms then followed before the onset of thesecond image for 1,200 ms. The blank frame then reappeared until theparticipant responded. On half of the trials, the second image of each paircontained one extra object; these constituted trials in the onset condition.On the other half of the trials, one of the already present old objectsunderwent a luminance change. For every trial type in which an additionalobject appeared on the second image (i.e., an onset trial), the same trialwould be repeated (i.e., occur twice in the experiment), with the exceptionthat the object would now be present in both images, and this object wouldchange luminance on the second image. In other words, an onset object onone trial would become a luminance change on another, and vice versa.This control ensured that neither onset nor luminance change had an unfairadvantage in localization performance over the whole experiment. Thus, if,for example, an onset had an advantage on one trial by virtue of, say, itslocation nearer to fixation or being a more salient color, this object wouldundergo a luminance change on another trial. This method of equating theperceptibility of onsets with that of the other stimulus attributes beingassessed in the present research was applied to all experiments reportedhere. Participants were asked to indicate whether they detected a change oneither the right or left side of fixation by pressing a right-hand or left-handbutton. If a change was not detected, the participant was asked to guess.The beginning of a trial was initiated by the participant’s response on the

previous trial. Participants were instructed to maintain fixation for theentire duration of each trial and that accuracy, rather than speed, wasimportant. They were not told what type of change to expect, only thatsomething would change across the two images. Participants were seatedapproximately 80 cm from the display. Onsets and luminance changesoccurred equally on the right and left sides of the display. Twenty practicetrials were given following a demonstration trial. Sixty-four onset trials and64 luminance change trials were presented in the experiment. All trial typeswere presented randomly.

Results and Discussion

Correct localization rates for the 64 onset trials were comparedwith those for the 64 luminance change trials. Mean scores were52.2 (82%) in the onset condition and 37.3 (58%) in the luminancechange condition. Participants’ mean scores for each of the twoconditions were entered into a within-participant t test. This re-vealed a significant difference in localization rates for the onsetand luminance conditions, t(9) � 7.77, p � .001. Thus, luminancechange at an old object was not as effective in localizing a changeas the appearance of a new object. The results therefore supportfindings showing that luminance change does not necessarilyattract attention (e.g., Enns et al., 2001; Yantis & Gibson, 1994;Yantis & Hillstrom, 1994). This provides initial evidence that theonset advantage observed by Cole et al. (2003) is an advantage forthe onset of an object. We do not, of course, suggest that lumi-nance processing plays no role in the orienting of attention to newobjects; clearly luminance processing is an effective mechanismfor alerting the system to a new object. We do however argue thatluminance detection is not a necessary condition of attentionalorientation.

Experiment 2

An abundance of physiological studies have provided evidenceof specific pathways responsible for color vision. For instance, the

Figure 3. The trial sequence for Experiment 1. In this example, an additional object is presented in the secondpair of images on the right-hand side. This constitutes an object-onset trial.


receptive fields of cells constituting the subcortical parvocellularstream are known to be color opponent, and monkeys who havehad lesions to this stream are not able to perform color discrimi-nation tasks (Schiller, Logothetis, & Charles, 1990). Furthermore,there is evidence of separate pathways for color and luminancedetection (Cavanagh, Anstis, & MacLeod, 1987; Gregory, 1977;Mullen & Losada, 1994). This makes adaptive sense because colorprovides an efficient way of parsing the visual scene. One mighttherefore also expect color vision to be involved in the detection ofnew objects.

However, as reviewed in the introduction, behavioral evidencedemonstrating the orientation of attention by a color singleton hasbeen inconclusive. Although a red line clearly “pops out” amongan array of green lines when color is task relevant, the issue iswhether color can direct exogenous attention in the absence of arelevant attentional set. Jonides and Yantis (1988) initially showedthat a target letter that also happened to be of a different color thandistractors did not summon automatic attention. However, Turattoand Galfano (2001) and Johnson et al. (2001) have more recentlyprovided evidence to the contrary. The purpose of Experiment 2,therefore, was to address the issue of whether the onset advantageis mediated by the detection of color change that accompaniesonset. More specifically, we assessed whether a color change of anold object would be as effective as the onset of a new object inlocalization performance. If the onset effect is mediated by mech-anisms that detect a new perceptual object, rather than low-levelchrominance change, we should again expect higher localizationperformance for onsets. An important aspect of the design was thatthe degree of color change that occurred at an old object was equal(in Lu’v’ color space) to that which occurred at the location of thenew onset.

Method


Stimuli and apparatus. All aspects of the stimuli and apparatus werethe same as in Experiment 1, with the following exceptions. For everyonset trial, a color condition trial occurred in which an old object changedcolor. The color change of the old objects was equal to that resulting fromthe onset of the new object. As in Experiment 1, this was measured inLu’v’ color space (whose color axes are designed so that distance equatesto discriminability). The average Lu’v’ distance of the color change was0.12 Lu’v’ units. Furthermore, all of the objects and background were ofequal luminance (76 cd/m2).

Design and procedure. All aspects of the experiment were as describedfor Experiment 1. Hence, a within-participant, single-variable (onset vs.color change), two-alternative forced-choice procedure was used. Partici-pants were asked to indicate whether they observed a change on theleft-hand or the right-hand side of the display and asked to guess if nochange was seen.


Mean localization scores for the onset and color change condi-tions were 48.5 (76%) and 36.0 (56%), respectively. This differ-ence of 12.5 proved to be significant, t(9) � 6.20, p � .001. Thedata therefore show that old objects undergoing a color changewere not as efficiently localized as new objects. This was observeddespite the old object color change being equivalent to that whichaccompanied a new onset. Although the data do not allow us to

conclude how effective color per se is in orienting attention, theydo allow us to conclude that the appearance of a new object hasprimacy over color change of an old object. This provides addi-tional evidence that the onset advantage is a result of object onsetrather than the detection of a low-level property. In addition, itmight appear that comparing onsets with luminance and/orchrominance change in a change detection task gives the onset anunfair advantage, the rationale being that any luminance or colorchange would be masked by the global luminance and colorchange that results from the blank frame. However, onsets are alsosubject to the same disruption created by the blank frame. Putanother way, onsets appear to be less susceptible to the distractingeffect created by a global transient than are other types of stimuluschange.

Experiment 3

The finding that neither luminance change nor chrominancechange is as effective as onset in localizing change suggests thatsuch low-level attributes are not necessary for enhanced perfor-mance. If this is the case, then a new object should confer the sameonset advantage observed in Experiments 1 and 2 when it and allcompeting old objects (i.e., the whole display) are defined byneither luminance nor chrominance. The aim of Experiment 3 wasto test this hypothesis by comparing onsets with offsets. Thus,rather than attempting to equate objects for luminance and color,we investigated the onset advantage effect with stimuli that pos-sessed neither of these two attributes. To this end, we used stimulidefined by motion direction. With this procedure, the backgroundwas made up of lines moving in one direction, and the objectscomprised lines moving in the opposite direction (see Figure 4).This resulted in the appearance of second-order contour at thelocation in the display where the lines moved in opposite direc-tions. Such contours can be used to create the subjective appear-ance of objects. Although luminance change does occur with thismethod (by virtue of pixels changing from black to white, and viceversa), there is continuous and equal luminance change at all partsof the screen, and no additional changes are introduced by objectonsets or offsets. Such stimuli can therefore be used to assessattentional effects when all low-level factors need to be eliminated.Hence, Gellatly et al. (1999) previously used motion-direction-defined displays to show that the new object advantage in the onsetsingleton task is not due to forward masking of placeholders.

The use of such stimuli also circumvents a potential confoundthat may have occurred in Experiments 1 and 2. Our method ofequating objects for luminance would not in practice have guar-anteed equiluminance. This is because equiluminance can be de-fined either as on-screen equivalence, retinal equivalence, orpostretinal neural response equivalence (Cavanagh, Adelson, &Heard, 1992). Clearly, therefore, our on-screen measurement ofequivalence might not have been represented as equivalent higherup the visual system. Furthermore, our use of Lu’v’ color space inExperiment 2 may also not have guaranteed equal discriminabilityof colors. Lu’v’ color space is a standard transformation of Com-mission Internationale de l’Eclairage space that was designed to beperceptually uniform. Discriminability between any pair of colorsshould be identical, provided the two colors are separated by thesame distance in Lu’v’ space. Nevertheless, it is widely acceptedthat even under optimal conditions of adaptation, Lu’v’ and related


color spaces are far from perfect in achieving the goal of uniformdiscriminability (Fairchild, 1998). Unless one psychophysicallyadjusts the colors of every pair of stimuli used in an experiment,the possibility remains that there may be some, albeit small,differences in their discriminability. Finally, onset and luminancechange in Experiment 1 may not have been truly perceptuallyequivalent given Weber’s law.

Method


Stimuli and apparatus. The stimuli were as shown in Figure 4. Thelines comprised gray pixels measuring one pixel in thickness (0.034°),spaced the same thickness apart and presented against a white background.Both upward and downward moving lines traveled at 2.3°/s. Unlike inExperiments 1 and 2, either 8, 9, or 10 objects would result from a newobject appearing, and either 7, 8, or 9 objects would result from an objectoffsetting. Objects were not equally divided between the left and right sidesof the display. Hence, there might occur 6 objects on one side and 3 on theother. Also, unlike in Experiments 1 and 2, no objects occluded otherobjects. Objects were sized between 0.7° and 2.1° in both width and height.

Design and procedure. A within-participant, single-variable (onset vs.offset), two-alternative forced-choice procedure was used. The secondimage of each pair contained either one extra object or one object less,constituting the object onset and object offset conditions, respectively. Forevery onset trial, the same pair of images would be repeated on anothertrial, with the presentation order of the two frames reversed. Thus, anyspecific onsetting object would also, on another trial of the experiment, bean offsetting object. This ensured that any difference in localization per-formance could be due to only the onset or offset of an object, with all otherfactors, such as eccentricity of the change location, equivalent across onset

and offset conditions. All other aspects of the design and procedure wereas described for Experiment 1.


For the 64 onset and 64 offset trials, participants responded witha mean correct response rate of 50.7 (79%) for onsets and 45.7(71%) for offsets. This difference proved to be significant, t(9) �3.35, p � .01. Experiment 3 therefore demonstrated the primacy ofonsets over offsets when neither was accompanied by change inluminance or chrominance. The increased efficiency with whichonsets are detected in the change detection paradigm is not, there-fore, the result of luminance or chrominance detection. The findingthat the onset advantage is not dependent on these low-level objectproperties suggests that the effect is as a result of mechanisms thatdetect a perceptual object. The aim of Experiments 4a and 4b wasto examine this hypothesis by presenting onsets that did not resultin the appearance of a new object.

Experiment 4a

As did the present Experiment 3, Watson and Humphreys(1995) assessed the relative efficiency with which onsets andoffsets are detected. Their participants were required to detect atarget letter among distractors. The crucial manipulation was howthe letters were created: either by the onset or by the offset of partsof existing objects. This contrasts with the usual onset singletonprocedure, in which all of the elements of one item (i.e., the newobject) onset. This allowed the authors to assess priority of whatthey referred to as contour onsets and contour offsets. Watson and

Figure 4. The motion-direction-defined stimuli used in Experiment 3. The objects comprise lines moving in anupward direction against a background of downward-moving lines.


Humphreys found that RT to detect a target was not dependent onhow the target had been created. In other words, no advantage wasobserved for onsets over offsets. This led the authors to claim thatthere is “no special role for onsets.” However, their procedure andsubsequent findings lead to an intriguing hypothesis. In Watsonand Humphreys’s experiments, onsets and offsets occurred toalready existing old objects. This contrasts with our method ofonsetting–offsetting complete, or whole, objects. Therefore, theonset advantage may occur only for the onset of objects per serather than the onset of any particular stimulus. As Watson andHumphreys (1995) surmised, “If, in previous studies, onsets cap-tured attention over offsets because onsets alone formed newobjects, onsets that do not also result in the formation of a newobject should not take priority over offset stimuli” (p. 584). Theaim of Experiment 4a was to address this issue.

In order to investigate this, we compared the onset and offset ofwhole objects with onset and offset of part of an object. Animportant aspect of the design in these partial-onset, partial-offsetconditions was the use of onset–offset stimuli that we alreadyknew could induce an onset advantage. If these changes (i.e.,partial onset or offset) were themselves undetectable (e.g., if theywere not salient enough), then any reduction observed in changedetection for partial onset–offset might simply be due to theinability of the stimulus change to be detected anyway, regardlessof whether it was part of another object or not. We would thereforenot be able to make any conclusions concerning any absence ofchange detection to a partial onset–offset. Thus, we comparedonsets and offsets of whole objects with a condition in which thesesame trials were repeated with the exception that the onsets–offsets were now part of one of the old objects (see Figure 5).Critically, therefore, the stimulus change that occurred for partialonsets–offsets was identical to that which occurred for wholeonsets–offsets.

Method

Participants. Sixteen undergraduate psychology students took part inpartial fulfillment of a course requirement.

Stimuli and apparatus. All aspects of the stimuli and apparatus werethe same as in Experiment 1, with the sole exception that all of the objectspresented in the display were squares and rectangles.2 A partial onsetincreased the size of an old object by approximately 100%, whereas apartial offset decreased it by 50%.

Design and procedure. A 2 � 2 fully repeated measures design wasused. The first factor manipulated whether the onset or offset occurred inisolation or was part of an old object (i.e., either the appearance ordisappearance of a complete object or part of an object). The second factormanipulated the type of change that occurred (i.e., onset or offset). As inExperiment 3, for every trial on which an onset occurred, the same pair ofimages would be repeated, with the presentation order of the two imagesreversed so that the onset would also become an offset. Sixty-four trialswere presented in each of the four conditions, making a total of 256 trialsin the experiment. All other aspects of the design were as described forExperiment 1.


Figure 6 shows mean detection rates for onsets and offsets inboth the partial onset–offset and whole onset–offset conditions.Each participant’s mean localization rate for each of the fourconditions was entered into a 2 � 2 within-participant analysis of

variance (ANOVA). Significant main effects were observed forwhole object changes compared with partial object changes, F(1,15) � 12.8, p � .01, as well as for change type, F(1, 15) � 24.4,p � .01. However, the most interesting result concerns the signif-icant Whole–Partial Change � Onset–Offset interaction, F(1,15) � 9.9, p � .01. The onset advantage was significantly reducedif the onset occurred as part of an already present object. Wecarried out a simple comparison using the error variance from theoverall ANOVA to assess whether onset was more readily detectedthan offset when each change was part of an old object. This effectfailed to reach conventional statistical significance, t(15) � 1.90,p � .10.

Overall, the results from Experiment 4a show that the onset–offset asymmetry observed in Experiment 3 and by Cole et al.(2003) is abolished if onset–offset occurs as part of an old object.This suggests that onsets that do not correspond with the appear-ance of a new perceptual object are not as effective in orientingattention. That is, the onset advantage occurs for object onsetrather than onset per se. This concurs with Watson and Hum-phreys’s (1995) hypothesis that “onsets that do not also result inthe formation of a new object should not take priority over offsetstimuli” (p. 584). This is central to the contention that the visualsystem assigns high priority to the appearance of new objects.

Experiment 4b

We contend that the abolishment of the onset advantage forpartial onsets in Experiment 4a occurred because this type of onsetdid not accompany the appearance of a new object. However, analternative explanation can be posited. Localization of partialonsets may have been reduced not because such onsets did notform a new object but because they occurred in close proximity toan already present object. In other words, the adjacent object, anobject of the same size and color, may have masked the onset. Theaim of Experiment 4b was to assess this possibility. We repeatedthe partial-onset condition of Experiment 4a and compared local-ization performance with that in a nearly identical condition inwhich the same trials were presented, with the sole exception thata small gap was introduced between the two proximate items (seeFigure 7). Hence, the amount of masking resulting from the items’close proximity was (virtually) the same in both conditions; in onecondition, however, an old item changed size, whereas in the othercondition, a new object appeared. If close proximity does indeedexplain the abolishment of the onset advantage in Experiment 4a,localization performance should be the same in both conditions. If,in contrast, proximity of the two items did not disrupt performancein Experiment 4a, and increased performance for onsets was due tothe onset of an object, we should observe a detection advantage inthe object onset condition compared with the partial-onsetcondition.

2 As stated, the partial onset–offset that occurred to an old object neededto be detectable on its own. This would have meant that the whole-objectonset–offset condition would have needed to include, for example, half-moon shapes or empty circle shapes (i.e., the stimulus shape created byadding to circle). Thus, for software reasons, it was easier to create squaresand rectangles.


Method

Participants. Twelve undergraduate psychology students took part inpartial fulfillment of a course requirement.

Stimuli and apparatus. All aspects of the stimuli in the partial onsetcondition were identical to those reported in Experiment 4a. Stimuli in theobject onset condition were also identical, with the sole exception that agap of 0.2° (or 6 pixels) occurred between the two closest edges of theonset and its adjacent object.

Design and procedure. A within-participant, single-variable (partialonset vs. object onset), two-alternative forced-choice procedure was used.

Sixty-four partial onset trials and 64 object onset trials were presented. Allother aspects of the experiment were as described for Experiment 1.


Mean detection scores for the object onset and partial onsetconditions were 46.4 (73%) and 41.8 (65%), respectively. Thisdifference proved to be significant, t(11) � 2.70, p � .05. Thesemeans are included in the figure showing the results from Exper-iment 4a (Figure 6). The data show that although the onset oc-

Figure 5. Examples of a whole-object onset trial (A) and a partial-object onset trial (B) presented inExperiment 4a. In both Panels A and B, an onset occurs in the bottom right-hand part of the display. Specificonset and offset trials occurred twice in the experiment, once in isolation as a whole object (A) and once as partof an object (B). Hence, the onset stimuli were identical in both conditions. This was also the case for the offsetconditions.


curred in equally close proximity to its adjacent item in bothconditions, the onset that created a new perceptual object was moreeffectively localized than the onset that merely altered an existingobject. This demonstrates that the abolishment of the onset advan-

tage in the partial-onset condition of Experiment 4a was not due tothe close proximity with which the onset occurred to the adjacentitem. This supports our contention that the advantage observed foronsets in the change detection paradigm is an advantage for thedetection of an object representation.

Experiment 5

The current Experiments 1–4b, and those of Cole et al. (2003),compared onsets with other stimulus attributes by analyzing de-tection rates on a number of onset trials with detection rates on anumber of trials on which the competing attribute occurred. Theaim of Experiments 5 and 6, however, was to investigate whetheronset would still have primacy when both an onset and a compet-ing attribute occurred within the same trial (i.e., when presentedsimultaneously). If object onset induces a genuine performanceadvantage over other stimulus properties, then onsets should bemore readily localized when they are in direct competition withother attributes. As Enns et al. (2001) noted, any strategic biasesthat may contribute to an observed effect should be eliminated iftwo attributes are compared within the same trial.

In Experiment 5, we pitted onsets against offsets. On each trialof the experiment, an onset occurred on one side of the display andan offset occurred on the other. This enabled us to assess theonset-versus-offset issue using this more stringent method. Previ-ous work examining the onset–offset issue has typically investi-gated the effectiveness with which different types of cue directattention in Posner’s (1980) precuing paradigm. In a variation ofPosner’s now-classic procedure, participants are presented with anonsetting cue that appears at a peripheral location prior to the onsetof a simple target. The crucial manipulation is the spatial relation-ship between the cue and target. When cue and target are in closespatial proximity to each other (valid cues), RT to detect the target

Figure 6. Results of Experiments 4a and 4b. Percentages of correctlocalization rates are plotted for objects onsetting–offsetting either inisolation or as part of another object. Absolute scores out of 64 andstandard deviations are shown in parentheses. The data shown with aster-isks on the right-hand side of the figure are detection rates from Experi-ment 4b. These compare partial onsets with onsets of whole objects thatappeared in close proximity to an old object. The lower value of 65represents the former, whereas the upper value of 73 represents the latter.

Figure 7. The whole-object onset condition of Experiment 4b. At the bottom right-hand location, a complete(new) object has appeared adjacent to an old object. Trials of this type were compared with the trial shown inFigure 5B.


is reduced relative to when both are spatially separated (invalidcues). This demonstrates that the cue directs attention to thelocation of its appearance. The orienting of attention in this mannerhas been shown to occur with various types of cue. For instance,the onset of a dot (Nakayama & Mackeben, 1989; Steinman et al.,1995), a square (von Grunau & Faubert, 1994), and a shorthorizontal line (Krose & Julesz, 1989) will all accrue attention. AsCole, Gellatly, and Blurton (2001) have noted, cues of this type,although this is rarely stated explicitly, can be considered newobject onsets and will be treated as such by the visual system.Assessing the effectiveness of onset and offset of this type of cueprovides a direct test of the hypothesis that the visual system isparticularly sensitive to object onset.

The work that has been undertaken in this area tends to suggestno advantage for either onset or offset in directing attention. Prattand McAuliffe (2001) manipulated the cuing procedure such thatthe target was preceded by a small disk that would either onset oroffset. Results showed that although the usual RT difference wasobserved for valid versus invalid cuing, no difference in RT wasshown for onset and offset cues. Theeuwes (1991) has also pro-vided data on the efficacy with which onset and offset cuessummon attention. In his experiments, participants performed astandard visual-search task in which they had to detect a targetletter among distracting letters. The cue would be either the onsetor the offset of a bar marker that appeared adjacent to one of thesearch items. Results also showed that RT to detect the target letterwas not dependent on the type of cue presented. Riggio, Bello, andUmilta (1998) also failed to observe an onset–offset cue asymme-try using a method similar to that of Pratt and McAuliffe (2001).

In the course of assessing attentional set effects (for onset–offset targets) induced by onset and offset cues, Atchley, Kramer,and Hillstrom (2000) also provided data on the relative efficacy ofonsets and offsets in orienting attention. As with the onset single-ton task, participants searched for a target letter among distractorscreated either by onsetting or offsetting elements. Prior to the onsetof the search display, either an onset or an offset cue appearedadjacent to one of the letters. In three experiments using this basicprocedure, the onset–offset cue effect was found to be mixed. Asmall (13-ms) advantage for offsets was observed in Atchley etal.’s first experiment, no difference was found in their secondexperiment, and an 8-ms onset advantage occurred in their thirdexperiment. When errors were analyzed, no differences in cue typewere found. Contrary to the new-object hypothesis, these studiessuggest that the onset of an object is no more effective in attractingattention than offset.

Method


Stimuli and apparatus. All stimulus properties and apparatus were asdescribed for Experiment 1.

Design and procedure. A single-factor (onset vs. offset) within-participant design was used. Frame durations were as described for Exper-iment 1. The second image of each pair contained one extra object on oneside of the display and one less object on the other side (i.e., an onset andan offset). Although two changes occurred on each trial, participants wereactually told that the “majority” of trials would include a single change,whereas on a “small minority” of trials, two changes would occur, one oneither side of fixation. Participants were asked to indicate if they detected

a change on either the right or the left side by pressing a right-hand or aleft-hand button. They were asked to respond left or right only if they weresure that they had detected a change. If they either failed to detect a changeor detected two changes, they were told to press the space bar. Askingparticipants to respond only when they had detected a change contrastswith Experiments 1–4b, in which participants were invited to guess. Forevery trial on which an onset occurred on one side and an offset occurredon the other, the same pair of images would be repeated, with the presen-tation order of the two images reversed. In other words, an offset object onone trial would become an onset object on another, and vice versa. As withthe other experiments reported here, this control ensured that neither onsetsnor offsets had an unfair advantage in localization performance over thewhole experiment. Twenty practice trials were given following a demon-stration trial. One hundred and twenty-eight trials were then presented inthe experiment. All trial types were presented randomly.


Participants responded on the side corresponding to an onset ona mean of 73.9 (58%) and on the side corresponding to an offseton 51.6 (40%) of trials. After the experiment was complete, eachparticipant was asked whether he or she had ever seen the twochanges. No participants reported seeing both changes on any ofthe trials. Thus, participants’ space-bar responses on 2% of trialsmust have been due to their failing to detect any change on thosetrials. Each participant’s mean score for onsets and offsets wasentered into a within-participant t test. The difference of 22.3proved to be significant, t(9) � 3.46, p � .01. The results thereforesupport those of Cole et al. (2003) and the present Experiments 3and 4a. The appearance of an object is more likely to be detectedthan its disappearance. This effect occurs when both onsets andoffsets are in direct competition with each other in the same trial.

The results do not, however, support previous studies showingan absence of onset–offset asymmetry (e.g., Pratt & McAuliffe,2001; Riggio et al., 1998; Theeuwes, 1991). We have alreadycommented on how these studies are at odds with the notion thatonsets are particularly effective in orienting attention. How can wereconcile these contrasting results? Apart from the obvious meth-odological differences in designs (i.e., our change detection taskand other researchers’ standard search task), we believe that amore fundamental reason has led to these studies failing to observean onset advantage. It is probable that previous onsets and offsetswere equally effective in capturing attention because of their statusas sole visual transients. Recall that, for instance, in the Theeuwes(1991) procedure, participants were presented with an attention-capturing cue prior to the onset of a search display containingletters. The cue was either the onset or offset of a bar marker.Crucially, when the onset or offset (cue) was introduced, no otherstimulus event occurred. It is well known that a visual transient, inthe absence of any other (competing) simultaneous event, willhave a high probability of attracting attention, regardless of whatkind of change it is. Indeed, change blindness studies use compet-ing visual transients precisely to avoid the change being detectedevery time. In effect, Theeuwes’s onset–offset cue conditions wereequivalent to a mudsplash change blindness trial without themudsplash. Such a procedure would give rise to near 100% detec-tion, regardless of whether the change was an onset or offset (oranything else). However, we believe that the general difficulty ofour experimental task (i.e., change detection) means that themethod is more sensitive to any onset–offset differences.


Experiment 6

The aim of Experiment 6 was to test another of our earlierclaims (from Experiment 2) with the method used in Experiment 5.We assessed whether an onset would again be subject to increasedlocalization performance when it was in direct competition with anequivalent color change that occurred within the same trial. Aswith Experiment 5, an onset occurred on one side of the displaysimultaneously with an old object changing color on the other side.Hence, all aspects of the experiment were identical to those ofExperiment 5, with the exception that one of the old objectschanged color (rather than offsetting) simultaneously with theonset of the new object.

Method


Stimuli and apparatus. All aspects of these were identical to those ofExperiment 2. Hence, the degree of color change occurring at an old objectwas equal to that occurring at the location of the new onset.

Design and procedure. All aspects of these were identical to those ofExperiment 5, with the exception that within each trial, an onset was pittedagainst an old object undergoing an equivalent color change. Also, 64experimental trials were presented rather than 128.


Of the 64 trials, participants localized an onset on a mean of27.6 (43%) trials and a color change on 8.8 (14%) trials. Partici-pants failed to detect any change on 27.5 (43%) trials. Similar toExperiment 5, no participant reported seeing both changes. Eachparticipant’s mean score for onsets and offsets was entered into awithin-participant t test. The difference of 18.8 proved to besignificant, t(9) � 5.50, p � .001. These results therefore showagain that an onset is more effectively localized than an equivalentcolor change when both are in direct competition with each other.One striking difference between these results and those of Exper-iment 5 is the reduction in overall localization performance. Forinstance, participants failed to detect a change on only 2% of trialsin Experiment 5. In contrast, this figure was 43% in Experiment 6.In a similar manner, onsets were correctly localized on 58% oftrials in Experiment 5 and 43% in Experiment 6. This is unsur-prising given the findings of Experiments 1–3; luminance andcolor information, which were controlled for in Experiment 6 butnot in Experiment 5, aid the onset advantage (but do not cause it).

Experiment 7

Rensink (2000) has argued that the change detection paradigm,in addition to revealing the limitations of perception, can serve asa useful technique for investigating other aspects of visual pro-cessing. Indeed, the present study is based on this rationale. If thisis the case, then the advantage for new objects reported here, aswell as for other visual phenomena potentially indexed by changedetection tasks, should occur independently of how one interfereswith the retinal transient that accompanies a scene change. Clearly,if the onset primacy effect only occurs when a particular method ofchange detection is used, the effect might only be due to someaspect of the procedure rather than a mechanism of the visualsystem. In our final experiment, we aimed to demonstrate the

efficacy of onsets compared with offsets when we used one othermethod of disrupting the retinal transient that accompanies onset.In Experiments 1–6, we used a blank frame between images toachieve this. In Experiment 7, however, we used the mudsplashtechnique, in which the image remains throughout the trial and anumber of small mudsplashes appear briefly and simultaneouslywith the change (O’Regan et al., 1999). The experiment wastherefore a direct replication of Cole et al.’s (2003, Experiment 2)procedure, with the sole exception that a mudsplash acted as theinterrupting transient rather than a uniformly blank frame.3

Method


Stimuli and apparatus. All stimulus properties were as reported for theonset condition in Experiment 1, with the exception that six black squaresoccurred at pseudorandom locations across the display instead of the blankframe. The black squares were equivalent to the mudsplashes typicallyused in this paradigm. No black square covered the location of the onset oroffset. The squares measured 1.3° along each side.

Design and procedure. A within-participant, single-variable (onset vs.offset), two-alternative forced-choice procedure was used. All aspects ofthe design and trial procedure for the onset condition were as described inExperiment 1, with the following exception. Rather than a blank framebeing presented, the objects and figures making up the display remainedon-screen, and the six small black squares appeared for 100 ms. Simulta-neous with the onset of the mudsplashes, the onset–offset occurred. Asdescribed previously, each specific trial would occur twice, once with anobject appearing and once with the same object offsetting. Hence, wereverted to the procedure whereby a number of onset trials (64) werecompared with the same number of offset trials (rather than an onset andoffset being presented within the same trial).


Mean scores for the onset and offset conditions were 49.7 (78%)and 45.6 (71%), respectively. This difference of 4.1 proved to besignificant, t(9) � 2.31, p � .05. Hence, the results demonstratethe selection priority of new objects when an additional method ofdisrupting the onset transient is used. This gives us confidence thatonsets are subject to a real processing advantage.

General Discussion

The principal aim of the present research was to address theissue of whether the visual system treats the appearance of a newobject with particular importance or whether other stimulus eventsare just as likely to attract visual attention. We adopted the one-shot change detection paradigm whereby detectability of newobjects was compared with a variety of changes that occurred atalready present, or old, objects. If object onset has a special role invisual attention, one might expect onsets to be less susceptible tochange blindness than other changes. For each trial of seven

3 Recall that the other onset-versus-offset experiments reported hereused variations of the basic paradigm (e.g., displays defined by motiondirection, stimuli controlled for luminance and color, and competingonsets–offsets presented within the same trial). Hence, the present exper-iment was identical to that of Cole et al. (2003, Experiment 2), with theexception that a mudsplash was used.


experiments, participants were presented with a single pair ofimages comprising simple shapes. The images were separated bya blank frame, or a mudsplash (Experiment 7), and participantsindicated whether a change occurred on the left-hand or the right-hand side of the display. Correct localization rates were thencompared for the different types of change.

Experiment 1 compared onsets with equivalent changes in lu-minance. Results showed that localization performance was betterfor onsets. In Experiment 2, onsets were compared with equivalentcolor changes. Again, localization performance was better foronsets. Experiment 3 compared onsets with offsets when all ob-jects in the display were defined by neither luminance nor chromi-nance but by motion direction. Thus, all low-level attributes of thestimuli were controlled. Again, we found primacy for onsets. InExperiment 4a, we investigated whether it is the detection of onsetper se that leads to enhanced localization performance or onset ofa whole object. Results showed that onset stimuli that produce anonset advantage will only do so when presented alone as singleobjects. When they were presented as part of an already existentold object, the onset advantage was eliminated. Experiment 4bshowed that this elimination was not caused by masking due to thepartial onset occurring in close proximity to an already presentobject. In Experiment 5, we pitted onsets directly against offsets bypresenting each within the same trial. Results showed that onsetswere again more readily localized. In a similar manner, Experi-ment 6 pitted onsets against equivalent color changes. Again,onsets were more frequently localized. Finally, in Experiment 7,we showed that in addition to occurring with the flicker procedure,the new-object advantage would occur when the retinal transientthat normally accompanies change is disrupted by a mudsplash.

Overall, these experiments demonstrated that the onset of a newobject attenuates the change blindness effect relative to otherstimulus events. Onsets were more readily localized than changesin luminance, changes in chrominance, or offsets. The finding thatequivalent changes in luminance and chrominance were not aseasily localized suggests that the onset advantage is not dependenton the detection of low-level stimulus properties or due to earlyvisual processes. This provides evidence that the effect was me-diated by mechanisms that detect the onset of a new perceptualobject, or that it was object based. The results from Experiment 4asupport this account; the onset advantage was eliminated for onsetsthat did not also result in the appearance of a new object.

The demonstration that the onset advantage occurred at the levelof object representation relates to one of the fundamental questionsin vision. This concerns the degree to which all stimulus input isprocessed before the system selects relevant units for higher pro-cessing. Space-based theories suggest that attention is allocated tounsegmented regions of space, with the common metaphor forsuch a system being an attentional spotlight (e.g., Broadbent,1982). Object-based theories, in contrast, propose that attention isdirected to objects and perceptual groups that are parsed preatten-tively on the basis of Gestalt principles of perceptual organization(e.g., Driver & Baylis, 1989; Duncan, 1984).

There is an abundance of evidence to support both space-basedand object-based theories of attention. An early demonstration ofthe spatial nature of attention was reported by Eriksen and Eriksen(1974). Using a response-competition paradigm, they showed thatthe efficiency with which a target could be detected was influencedby the spatial arrangement of irrelevant distractors. This suggests

that the restrictions imposed by a capacity-limited attention systemare spatial limitations. Evidence for object-based attention hascome from the general finding that unattended information isprocessed to a relatively high level. For instance, distractors maybe spatially distant from a target, but if they are grouped togetherwith the target—on the basis of, for example, common motion—their distracting effect may be as strong (Driver & Baylis, 1989;but see also Berry & Klein, 1993; Kramer, Tham, & Yeh, 1991).Evidence for object-based attention has also come from studies ofneurological patients. Driver and Halligan (1991) found that someneglect patients not only fail to attend to the left side of visualspace, they also neglect the left half of objects presented in theirright visual field. Hence, this suggests that the deficit is objectbased. Since the new-object advantage in the present experimentswas not dependent on the processing of low-level properties,object detection can be said to have occurred at a relatively highlevel of object description.

Previous research examining new-object priority in selectionhas primarily relied on the feature-singleton paradigm. More spe-cifically, selection priority has been demonstrated for new-objectsingletons, but it has been demonstrated less conclusively forcolor, luminance, or motion singletons. However, feature-singleton experiments may not be the most efficient way of dem-onstrating new-object priority. For instance, in such experiments,different types of singleton are usually compared with onset sin-gletons across experiments. We, however, compared onset directlywith a host of other changes within the same experiment, some-times within the same trial. Our method therefore allowed a directcomparison of object onset with a number of other stimulus events.This also means that object primacy did not rest on a null result.That is, evidence for selection priority of new objects has previ-ously relied on no effect being observed in other feature-singletonexperiments. Feature-singleton experiments may also confound theeffects of the unique stimulus and the global onset of the stimulusdisplay. Turatto and Galfano (2001) suggested that a color single-ton may well be able to orient attention, but its potential effect indirecting attention may be overridden by the global onset of thesearch display. That is, the signal from the singleton could bemasked by the signal from the onset of all of the items. We believe,therefore, that the change detection paradigm provides an alterna-tive method for assessing the new-object-priority hypothesis.

A major aspect of the present experiments was our attempt toequate discriminability of the different types of changes. All low-level information present in object onset was equal to that presentin luminance, chrominance, offset, and partial-object changes.This is a crucial procedural consideration for comparison of thedetectability of different properties in change detection. For in-stance, Rensink, O’Regan, and Clark (1997) examined whetherchange blindness occurs as a result of the absence of focusedattention at the location of change. They compared changes thatoccurred to objects of central interest with those that occurred toobjects of marginal interest. Results showed that changes of centralinterest were detected more rapidly than changes of marginalinterest. Although the central–marginal interest manipulation oc-curred for changes in location, color, and presence–absence, Ren-sink et al. were not able to compare RTs across these types ofchange. That is, comparisons could only be made for changes ofcentral–marginal interest within each of the three types of change,not between them. As Rensink et al. (1997) stated, “Because


discriminability was not equated for the three different types ofchange, performance between them cannot be compared” (p. 370).We are unaware of any previous change detection work that hasenabled direct comparisons between different changes to be madeby systematically controlling for change discriminability.

An interesting aspect of the current data concerns the magnitudeof the onset advantage across the different experiments. Althoughwe did not test this for statistical significance, onset primacyappears to have progressively reduced in Experiments 1–3. Whenwe attempted to control for luminance only (Experiment 1), onsetsenjoyed a 24% advantage in localization. When we attempted tocontrol for both luminance and chrominance (Experiment 2), thisadvantage was reduced to 20%. When luminance and chrominancefactors were completely eliminated (Experiment 3), the onsetadvantage was reduced to 8%. This suggests that luminance andchrominance change did in fact contribute to the onset advantageobserved in Experiment 1. Crucially, however, their contributionsonly added to the effect, they did not cause it. This is unsurprisinggiven that luminance transients, for instance, are effective inorienting attention. Finally, the onset advantage was reduced toonly 7% when a mudsplash, rather than a blank frame, was used todisrupt the change transient (Experiment 7). This clearly suggeststhat mudsplashes do not as effectively compete with the changetransient for attention as does a blank frame.

The question remains as to why the onset of an object morereadily attenuates change blindness compared with other objectchanges. Yantis and Hillstrom (1994) invoked Kahneman andTreisman’s (1984) notion of object files to suggest why onsetshave primacy in selection. An object file is “the process of settingup and utilizing temporary ‘episodic’ representations of real worldobjects” (Kahneman, Treisman, & Gibbs, 1992, p. 176). Theserepresentations include information concerning all of the attributesof an object, such as its location and its color. When one of theseattributes changes, the object file only needs to be updated. How-ever, when a new object appears, a new object file has to becreated. It is the creation of this new file that directs attention to theonset.

In summary, the experiments reported here show that objectonset attenuates change blindness relative to a variety of otherchanges. This suggests that the appearance of a new object is ofparticular importance in visual attention. Object onset does indeedappear to play a “special role” in visual attention.

References

Atchley, P., Kramer, A. F., & Hillstrom, A. P. (2000). Contingent capturefor onsets and offsets: Attentional set for perceptual transients. Journalof Experimental Psychology: Human Perception and Performance, 26,594–606.

Berry, G., & Klein, R. M. (1993). Does motion-induced grouping modulatethe flanker compatibility effect: A failure to replicate Driver & Baylis.Canadian Journal of Experimental Psychology, 47, 714–729.

Broadbent, D. E. (1982). Task combination and selective intake of infor-mation. Acta Psychologia, 50, 253–290.

Cavanagh, P., Adelson, E. H., & Heard, P. (1992). Vision with equilumi-nant colour contrast: A large scale technique and observations. Percep-tion, 21, 219–226.

Cavanagh, P., Anstis, S. M., & MacLeod, D. I. A. (1987). Equiluminance:Spatial and temporal factors and the contribution of blue-sensitive cones.Journal of the Optical Society of America A, 4, 1428–1438.

Cole, G. G., Gellatly, A. R. H., & Blurton, A. (2001). The effect of objectonset on the distribution of visual attention. Journal of ExperimentalPsychology: Human Perception and Performance, 27, 1356–1368.

Cole, G. G., Kentridge, R., Gellatly, A. R. H., & Heywood, C. (2003).Dectectability of onsets versus offsets in the change detection paradigm.Journal of Vision, 3, 22–31.

Driver, J., & Baylis, G. C. (1989). Movement and visual attention: Thespotlight metaphor breaks down. Journal of Experimental Psychology:Human Perception and Performance, 15, 448–456.

Driver, J., & Halligan, P. W. (1991). Can visual neglect operate in object-centered co-ordinates?: An affirmative single-case study. Cognitive Neu-ropsychology, 8, 475–496.

Duncan, J. (1984). Selective attention and the organization of visualinformation. Journal of Experimental Psychology: General, 113, 501–517.

Enns, J. T., Austen, E. L., Di Lollo, V., Rauschenberger, R., & Yantis, S.(2001). New objects dominate luminance transients in setting attentionalpriority. Journal of Experimental Psychology: Human Perception andPerformance, 27, 1287–1302.

Eriksen, B. A., & Eriksen, C. W. (1974). Effects of noise letters upon theidentification of a target letter in a nonsearch task. Perception & Psy-chophysics, 16, 143–149.

Fairchild, M. D. (1998). Color appearance models. Reading, MA:Addison-Wesley.

Folk, C. L., Remington, R. W., & Johnston, J. C. (1992). Involuntarycovert orienting is contingent on attentional control settings. Journal ofExperimental Psychology: Human Perception and Performance, 18,1030–1044.

Gellatly, A. R. H., & Cole, G. G. (2000). Accuracy of target detection innew-object and old-object displays. Journal of Experimental Psychol-ogy: Human Perception and Performance, 26, 889–899.

Gellatly, A. R. H., Cole, G. G., & Blurton, A. (1999). Do equiluminantobject onsets capture visual attention? Journal of Experimental Psychol-ogy: Human Perception and Performance, 25, 1609–1624.

Gibson, B. S., & Jiang, Y. (1998). Surprise! An unexpected color singletondoes not capture attention in visual search. Psychological Science, 9,176–182.

Gregory, R. L. (1977). Vision with isoluminant colour contrast: A projec-tion technique and observations. Perception, 6, 113–119.

Hillstrom, A. P., & Yantis, S. (1994). Visual motion and attentionalcapture. Perception & Psychophysics, 55, 399–411.

Horstmann, G. (2002). Evidence for attentional capture by a surprisingcolor singleton in visual search. Psychological Science, 13, 499–505.

Johnson, J. D., Hutchison, K. A., & Neill, W. T. (2001). Attentionalcapture by irrelevant color singletons. Journal of Experimental Psychol-ogy: Human Perception and Performance, 27, 841–847.

Jonides, J. (1981). Voluntary versus automatic control over the mind’seye’s movement. In J. B. Long & A. D. Baddeley (Eds.), Attention andperformance IX (pp. 187–203). Hillsdale, NJ: Erlbaum.

Jonides, J., Irwin, D. E., & Yantis, S. (1983, October 14). Failure tointegrate information from successive fixations. Science, 222, 188.

Jonides, J., & Yantis, S. (1988). Uniqueness of abrupt visual onset incapturing attention. Perception & Psychophysics, 43, 346–354.

Kahneman, D., & Treisman, A. (1984). Changing views of attention andautomaticity. In R. Parasuraman & D. R. Davies (Eds.), Varieties ofattention (pp. 29–61). Orlando, FL: Academic Press.

Kahneman, D., Treisman, A., & Gibbs, B. J. (1992). The reviewing ofobject files: Object-specific integration of information. Cognitive Psy-chology, 24, 175–219.

Kentridge, R. W., Cole, G. G., & Heywood, C. A. (2003). The primacy ofchromatic edge processing in normal and cerebrally achromatopsicsubjects. Progress in Brain Research, 144, 161–167.

Kramer, A. F., Tham, M., & Yeh, Y. (1991). Movement and focused


attention: A failure to replicate. Perception & Psychophysics, 50, 537–546.

Krose, B. J. A., & Julesz, B. (1989). The control and speed of shifts ofattention. Vision Research, 29, 1607–1619.

Martin-Emerson, R., & Kramer, A. F. (1997). Offset transients modulateattentional capture by sudden onsets. Perception & Psychophysics, 59,739–751.

Miller, J. (1989). The control of attention by abrupt visual onsets andoffsets. Perception & Psychophysics, 45, 567–571.

Mullen, K. T., & Losada, M. A. (1994). Evidence for separate pathways forcolor and luminance detection mechanisms. Journal of the OpticalSociety of America A, 11, 3136–3151.

Nakayama, K., & Mackeben, M. (1989). Sustained and transient compo-nents of focal visual attention. Vision Research, 29, 1631–1647.

O’Regan, J. K., Rensink, R. A., & Clark, J. J. (1999, March 4). Change-blindness as a result of “mudsplashes.” Nature, 398, 34.

Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Exper-imental Psychology, 32, 3–25.

Pratt, J., & McAuliffe, J. (2001). The effects of onset and offset on visualattention. Psychological Research/Psychologische Forschung, 65, 185–191.

Rayner, K., & Pollatsek, A. (1983). Is visual information integrated acrosssaccades? Perception & Psychophysics, 34, 39–48.

Rensink, R. A. (2000). Seeing, sensing, and scrutinizing. Vision Research,40, 1469–1487.

Rensink, R. A., O’Regan, J. K., & Clark, J. J. (1997). To see or not to see:The need for attention to perceive changes in scenes. PsychologicalScience, 8, 368–373.

Riggio, L., Bello, A., & Umilta, C. (1998). Inhibitory and facilitatoryeffects of cue onset and offset. Psychological Research/PsychologischeForschung, 61, 107–118.

Schiller, P. H., Logothetis, N. K., & Charles, E. R. (1990). Role of thecolor-opponent and broad-band channels in vision. Visual Neuroscience,5, 321–346.

Simons, D. J. (1996). In sight, out of mind: When object representationsfail. Psychological Science, 7, 301–305.

Simons, D. J., & Levin, D. T. (1997). Change blindness. Trends inCognitive Sciences, 1, 261–267.

Steinman, B. A., Steinman, S. B., & Lehmkuhle, S. (1995). Visual attentionmechanisms show a center-surround organization. Vision Research, 35,1859–1869.

Theeuwes, J. (1991). Exogenous and endogenous control of attention: Theeffect of visual onsets and offsets. Perception & Psychophysics, 49,83–90.

Theeuwes, J. (1994). Stimulus-driven capture and attentional set: Selectivesearch for color and visual abrupt onsets Journal of ExperimentalPsychology: Human Perception and Performance, 20, 799–806.

Titchener, E. B. (1901). Experimental psychology. New York: Macmillan.Turatto, M., & Galfano, G. (2001). Attentional capture by color without

any relevant attentional set. Perception & Psychophysics, 63, 286–297.von Grunau, M., & Faubert, J. (1994). Intraattribute and interattribute

motion induction. Perception, 23, 913–928.Watson, D. G., & Humphreys, G. W. (1995). Attention capture by contour

onsets and offsets: No special role for onsets. Perception & Psychophys-ics, 57, 583–597.

Yantis, S. (1998). The control of visual attention. In H. Pashler (Ed.),Attention (pp. 223–256). Hove, England: Psychology Press.

Yantis, S., & Gibson, B. S. (1994). Object continuity in apparent motionand attention. Canadian Journal of Psychology, 48, 182–204.

Yantis, S., & Hillstrom, A. P. (1994). Stimulus-driven attentional capture:Evidence from equiluminant visual objects. Journal of ExperimentalPsychology: Human Perception and Performance, 20, 95–107.

Yantis, S., & Johnson, D. N. (1990). Mechanisms of attentional priority.Journal of Experimental Psychology: Human Perception and Perfor-mance, 16, 812–825.

Yantis, S., & Jonides, J. (1984). Abrupt visual onsets and selective atten-tion: Evidence from visual search. Journal of Experimental Psychology:Human Perception and Performance, 10, 601–621.

Received August 9, 2002Revision received October 24, 2003

Accepted October 27, 2003 �


Visual Salience in the Change Detection Paradigm: The...

Documents

Transcript of Visual Salience in the Change Detection Paradigm: The...